Baidu upgrades its flagship AI models with reduced pricing

Baidu Inc. rolled out two faster Ernie models and cut their prices by as much as 80%, a step that lifted its Hong Kong-listed shares by more than 5%.

The tech firm introduced Ernie 4.5 Turbo, its new foundation model, and Ernie X1 Turbo, a reasoning model aimed to rival DeepSeek.

Billionaire founder Robin Li said during a developer meeting in Wuhan that 4.5 Turbo now costs only one-fifth of its earlier version, while X1 Turbo is priced at half of the old X1.

“All these releases are meant to let developers build apps without worrying about model power, cost or tools,” Li said.

Li also questioned the staying power of text-only systems such as DeepSeek’s R1. “The market for text models is shrinking,” he said, arguing that R1 is slower, more expensive and more prone to hallucinations than newer home-grown products.

Both Ernie updates handle not only words but also sound, pictures, and video, reflecting a wider push toward multimodal AI that can read, see, and listen in one interface.

Baidu showed an upgraded avatar tool that lets online sellers generate life-like hosts for livestream shopping and product demos, one of several examples aimed at business users.

AI competition in China is heating up

Competition in China has intensified since OpenAI’s ChatGPT shook the sector in November 2022. Baidu, the first Chinese company to answer ChatGPT, is now rivaling Alibaba, Huawei, Tencent, and a string of start-ups for corporate and consumer users.

The company has been forced to adapt. After a lukewarm response to its paid chatbot, Baidu dropped the subscription this year and opened much of its code to programmers, reversing an earlier stance that closed-source software was essential. Li said the field “changes every week” as fresh contenders release “powerful new models” that give buyers more choice.

In practice, Baidu still collaborates with its rivals. DeepSeek’s R1 is already offered on Baidu’s Qianfan cloud platform and inside its search and map services, signaling a pragmatic approach even as Li highlighted R1’s limits.

Industry analyst Charlie Dai of Forrester Research said Friday’s price cuts “will speed up AI use across Chinese industries, lower entry barriers for coders and tighten the race with Alibaba Cloud and others.”

At the same time, DeepSeek, backed by venture funds and praised for its reasoning skills, is working on R2 and V4 follow-ups, according to people familiar with its plans.

The Wuhan event also saw Baidu unveil Xinxiang, an AI agent app that will compete with Alibaba’s Quark and smaller newcomers such as Manus AI.

Backing the software, Baidu said it has built a computing cluster of 30,000 Kunlun P800 chips, made by its semiconductor arm. Li told the audience the cluster can train “several DeepSeek-like models at the same time,” so programmers “don’t need to worry about a shortfall” in processing power.

Last month, the Financial Times reported that Samsung had shipped Kunlun a three-year supply of key logic chips before new U.S. export rules threatened future deliveries, adding another tension point in the global tech supply chain.

Cryptopolitan Academy: Tired of market swings? Learn how DeFi can help you build steady passive income. Register Now